Classification Accuracy is Degrading

I am classifying text based on news headlines and I am achieving accuracy more than 80%. I want to improve it more.

But issue is that when I calculate the same with synonyms using the code below:

Doc = actxserver('Word.Application')    
X = cellfun(@(word) invoke( Doc,'SynonymInfo',word), words, 'UniformOutput', false);
Synonyms = cellfun(@(X) get(X,'MeaningList'), X, 'UniformOutput', false);
Synonyms = cellfun(@(X) [words{X}; Synonyms{X}], num2cell(1:numel(words)), 'UniformOutput', false);

My accuracy falls very badly below and reach 40% or less.

Why is this happening?

[José-Luis]

That would depend more on the text than on Matlab, or am I missing something?

[Mazhar Iqbal Rana]

Well! Text is divided in array of words already... So, synonyms are of words individually line by line

Answer 1

[Walter Roberson]

Because words can be used in different ways, you can have a word S that is a synonym of word A, and of word B, and yet word A and word B might not be synonyms. Especially if you happen to encounter words which are synonyms of one of the many meanings of the word "set" or "jack". For example, a "set" of cards or a "set" of dishes is a "collection", and to let glue "set" is to let it "cure", but "collection" and "cure" are not synonyms. You might have had "collection" and "cure" individually as being distinct, but when you add synonyms you add in "set" and that links the "collection" and "cure" and makes it more difficult to classify headlines that involve the words.

[Mazhar Iqbal Rana]

That is right Thanks a lot but what you suggest me for now. I intend to increase my classification accuracy and it is usually presumed that synonyms do so.. how could they in that case?

Thanks for your Kind support